Lo-Mendell-Rubin likelihood ratio test question?

[Tianfang Yang]

Dear developers of the tidyLPA package, 

Hope you are all staying well and healthy during the coronavirus outbreak! I have been trying to catch up with the latest update of tidyLPA, which incorporates the LMR LRT test through the calc_lrt() function. However, I found myself a bit lost when trying to run the LMR LRT to follow up a previous LPA on my personality dataset because I could not find information such as parameters and log-likelihood for both null and alternative models, which are necessary indices for running the calc_lrt() function as I understand. Should I look back into the tibble list output from my model to get all necessary information and is there a way to easily pipe the function into previous functions? 

Thank you very much!

Sincerely,

Tianfang

[rem...@gmail.com]

Hello Tianfang, 

Just responding because I have run into a similar issue when trying to apply the chi squared difference test to two models with the same number of classes but different parameterizations, as detailed in the article by Pastor and colleagues (2007). It does seem that the number of parameters, degrees of freedom, and LL values are in the tibble output for each model, which I have been saving as objects. I was able to calculate the LMR by calling the function calc_lrt() and then plugging in the values from the output. However, I have not figured out how to calculate the total parameter value for my own equation, which  entails  "taking two times the difference of the log-likelihoods of nested models and comparing the value against a chi squared distribution with degrees of freedom equal to the difference in the number of parameters being estimated". I will be following this thread closely for any other tips and best of luck! 

-Gary